WO2000031568A1

WO2000031568A1 - Processing well log data

Info

Publication number: WO2000031568A1
Application number: PCT/IB1998/001846
Authority: WO
Inventors: Anthony Smits; Vivian Pistre
Original assignee: Schlumberger Limited; Schlumberger Technology B.V.; Schlumberger Surenco S.A.; Petroleum Research And Development N.V.; Schlumberger Canada Limited; Schlumberger Overseas S.A.; Services Petroliers Schlumberger; Schlumberger Holdings Limited
Priority date: 1998-11-20
Filing date: 1998-11-20
Publication date: 2000-06-02
Also published as: AU1048799A; GB2359135B; GB2359135A; GB0111887D0

Abstract

Method of processing data obtained from an interval of a borehole using an array tool, such as a sonic logging tool, to make a series of measurements in the interval, the method comprising: (i) determining from data, a set of parameters, such as transit time differences or formation slowness, relating to the formation for each pair of receivers in the array for each measurement in the series; (ii) removing any parameters which fall outside predetermined limits based on the known physical properties of the borehole; (iii) comparing pairs of parameters relating to measurements at the same location in the interval; and (iv) statistical processing of the comparisons to determine an acceptable value for the parameter at each location in the interval.

Description

PROCESSING WELL LOG DATA

TECHNICAL FIELD

The present invention relates to the processing of well log data, and in particular to the processing of log data from an array of receivers to remove misleading or unrepresentative data from the dataset acquired from such an array.

BACKGROUND ART

Array logging tools for logging underground formations surrounding a borehole typically comprise one or more signal sources and an array of receivers. Such tools are often acoustic logging tools in which a transmitter excites an acoustic signal which passes through the formation to an array of receivers spaced from the transmitter. By measuring the time between detection of the signal at each receiver, it is possible to determine the speed at which the signal moves through the formation and hence the nature of the formation. The measure of movement of an acoustic signal through the formation is usually known as the "slowness" of the formation (the time for the signal to move a given distance, usually given in μs/ft). One example of such a tool is the DSI tool of Schlumberger which is described in more detail in US 4,850,450; US 4,862,991 ; US 4,872,526; US 5,036,945; and US 5,043,952.

A different type of logging tool to that described above is shown in Figure 1 , and comprises a pair of acoustic transmitters Tl , T2 and an array of five acoustic receivers Rl - R5 positioned half way between the two transmitters. For this tool, the manner of processing data is to consider each pair of receivers RxRy and to determine the difference in transit time for a signal from a given transmitter Tz to reach each of the receivers (ΔT _> = T_rzl - T _RΛ). The manner of detecting and determining ΔT is described in WO 97/28464 and the difference in transit times is assumed to be a measure of the formation slowness at the mid-point between the two receivers. If the array of receivers has sufficient members and is regularly spaced, this approach will lead to redundancy in the measurement since the array will give several measurements of the slowness at certain points as measured by different combinations of receivers. This redundancy is enhanced when two transmitters are used as in the tool of Figure 1. While a degree of variability is inevitable between measurements of the same region, from time to time, there will be measurements which have such large scale errors that including them in processing to determine formation properties leads to misleading or inaccurate results.

This problem can apply to different types of array measurements in boreholes and can be stated as follows (sonic logging issues given in parentheses): Given a set of transmitter to receiver (slowness) measurements from an array (sonic) tool, some of which may be affected by large scale errors (e.g. false detection), and others affected by small scale errors (e.g. ray path effects, depth errors, alteration effects), find the "best estimate " of the formation property (slowness) by rejecting the large scale errors and compensating, where possible, for the small scale errors.

The present invention seeks to provide a processing method for array well log data which reduces the impact of "bad" data on the determination of formation properties from that data.

DISCLOSURE OF INVENTION

In its broadest scope, the invention comprises a method of processing data obtained from an interval of a borehole using an array tool to make a series of measurements in the interval, the method comprising:

(i) determining from data, a set of parameters relating to the formation for each pair of receivers in the array for each measurement in the series;

(ii) comparing pairs of parameters relating to measurements at the same location in the interval; and

(iii) statistical processing of the comparisons to determine an acceptable value for the parameter at each location in the interval.

Preferably any parameters which fall outside predetermined limits based on the known physical properties of the borehole are removed prior to comparing the pairs of parameters.

There are two particular methods falling within the scope of the invention relating to processing sonic logging data. The first method includes the steps of:

(1) computing slowness curves for the interval for each pair of receivers in the array; (2) eliminating extreme values lying outside the predetermined physical thresholds;

(3) determining the separation of each curve from the other curves;

(4) forming groups of curves based on the determined separations;

(5) selecting the group(s) with the largest number of curves; (6) processing the group statistically to remove outliers; and

(7) determining formation properties from the processed group.

The second method includes steps of:

(1) determining transit time differences for each pair of receivers; (2) eliminating extreme values lying outside the predetermined physical thresholds; and

(3) determining the Hodges-Lehman average or the median of the remaining transit times to produce a slowness curve for the interval.

In both cases, computing curve separations or Hodges-Lehman average, the method of the invention looks at comparisons between members within the dataset to be analysed to attempt to identify the good data.

While not limited to the field of acoustic logging, this invention finds particular application to this area. A suitable tool comprises an array of receivers, for example five regularly spaced receivers, and one or more transmitters, for example two transmitters with the receiver array disposed between them. Such a tool can give two modes of formation slowness measurement: receiver mode and transmitter mode. Receiver mode slowness is the difference between transit times measured on a pair of receivers on the same firing of a transmitter, the measurement point being the mid-point between the two receivers. Transmitter mode slowness is the difference in transit times from a pair of receivers on different firings of the transmitter, the measurements being made when each receiver is at a given depth and the measurement point being the mid-point between the depths of the transmitter at each firing. By combining the different modes from different transmitters, it is possible to obtain a set of measurements which allow borehole and other interfering effects to be compensated. The preferred method of measuring transit times (the time for a signal to pass from the transmitter to a receiver) is to use the approach described in WO 97/28464 which will be called Digital First Arrival Detection or DFAD below. Other outputs from this approach can be used in the method of the invention to assist in determining good or bad data.

The present invention will be described below in relation to the accompanying drawings. It will be appreciated that changes may be made to this implementation while still falling within the scope of the invention.

BRIEF DESCRIPTION OF DRAWINGS

Figure 1 shows an acoustic borehole logging tool;

Figure 2 shows a plot of transit time TT against transmitter - receiver spacing TR;

Figures 3a and 3b show signal paths for a borehole through a homogenous formation and through a formation including a bed boundary; Figures 4 shows a high-level flow diagram of a method according to one embodiment of the invention;

Figure 5 shows a flow diagram of a data processing scheme incorporating methods within the scope of this invention;

Figure 6 shows a diagrammatic representation of groupings of slowness curves; Figure 7 shows a logic diagram for treatment of multiple slowness groups; and

Figure 8 shows a diagrammatic representation of using preference lists to compute outputs.

BEST MODE FOR CARRYING OUT THE INVENTION

Referring now to the drawings, the primary output from the tool of Figure 1 when processed using the DFAD algorithm is an estimated arrival time. Other outputs include a status word to indicate the operational state of the algorithm, the amplitude of the signal peak at the point of detection and a measurement of the noise in the signal before the detection window.

The receiver mode slowness is computed as the difference between the transit times measured on a pair of receivers on the same firing of the transmitters and is applied to the mid point between the receivers at that firing. The formation slowness ΔT(k) at depth index, k can be computed from transit times, 7T( j, at depth index, i, as

AT(k ) = [7T, ( ) - TT₂ (i)]/[rx₂ - rx, ]

k = i +

+ rx₂ )/2 + if transmitter above receivers

k - i + (rx, + rx₂)J2 -

if transmitter below receivers where rx, and rjc are the distances for receivers 1 and 2 from the transit time measure reference point, Δh is the depth offset due to ray angle, and Δz is the depth sampling interval. Figure 3(a) shows a schematic signal path for a receiver mode slowness measurement described above.

The transmitter mode slowness is computed as the difference in transit times from a pair of receivers on different firings of the transmitter. These different firings are selected such that the two receivers are at the same depth in the well. Thus the measure point of the resulting slowness corresponds to the mid-point of the transmitter positions for the two firings. The formation slowness ΔT(k), at depth index, k, can be computed from the transit time on receiver #1, ΔT,(i), at a depth index, /^', and transit time on receiver #2, ΔTfj), at depth index j, for a transmitter at a distance, tx, from the transit time reference point as

AT(k ) = [TT, ( ) - TT₂ {j)]/[rx₂ - rx, ]

k = i + Ytx + (rx, - rx₂)/2 - |Δ/?|J/|ΔZ| if transmitter above receivers

k

if transmitter below receivers

Measured transit times TT plotted against transmitter-receiver spacing TR for a homogenous formation should lie on a straight line through the origin. However, as is shown in Figure 2, a plot of actual measurements results in a small offset δTT. This offset can arise due to small errors in the measured transit time or due to the propagation time of the signal in the borehole fluid. For a tool in the situation shown in Figure 3(a), with a standoff from the borehole wall of x, in a fluid of slowness s_m, and a formation of slowness s, the TT intercept will be: όTT = 2[s_mx/cos( )- sx tan(φ)]

= 2x{ si - s²

For example, for a 18mm diameter receiver centred in a 200mm hole with a formation slowness s of 50μs/ft and a fluid slowness s_m of 200μs/ft, the offset will be 116μs. This is a more extreme case and the offset will usually be less than this.

Certain constraints can be placed on the processing from a knowledge of the physical behaviour of the borehole and formation. Obviously, the formation slowness cannot be less than Oμs/ft but practical experience has shown that the minimum slowness observed is of the order of 40 - 45μs/ft. Slownesses lower than this can usually be eliminated as errors. The system cannot measure a formation slowness that is slower than the borehole fluid slowness. This implies an upper limit of 180 - 200μs/ft for water based muds. In oil based muds the limit may be even higher.

One embodiment of the invention involves the Hodges-Lehman averaging of a group of slownesses, and comprises the following steps:

1. Compute the ΔT transit time difference (~ slowness) between each receiver pair separated by a given distance (for example, 6") in both receiver mode and transmitter mode.

2. Reject any slownesses less than or greater than a predefined minimum or maximum (for example, 40μs/ft and 180μs/ft). This will be sufficient in most cases to reject slownesses in which the transit time from the nearer of the pair of receivers has skipped onto a later peak which typically causes the nearer transit time to read high by at least 30μs and the computed slowness to fall below the lower limit. However, this may not be sufficient to pick up skips in transit time from the farther receiver in relatively fast formations because the increase in slowness might not be sufficient to exceed the upper limit. These limits may also not detect problems resulting from early detection on noise. 3. The offset in transit time oTT described above is applied to the remaining data and the upper and lower limits applied again as in 2 above.

4. The Hodges-Lehman average of the remaining data is calculated by computing the average values of all combinations of pairs and determining the median value of the result. (Alternatively, the median of the remaining data may be used)

5. The arrangement of the tool of Figure 1 can potentially give four measurements for each slowness and so allows a borehole compensated (BHC) result to be obtained if enough good data remains. If receiver mode measurements are available form both the upper and lower transmitter, the BHC slowness can be output. If one receiver mode is missing, either an uncompensated, or a BHC transmitter mode result can be output.

This approach is relatively simple and robust and so can be implemented in a relatively easy fashion without requiring excessive computing power. However, in cases where the formation is not homogenous or includes a bed boundary, a different approach may be required. Figure 3(b) shows the schematic signal path across a bed boundary with formations of slowness s₀ and s_r In this case, it is possible that the method described above will be unable to distinguish a skip on the near receiver in a slow formation, or on the far receiver in a fast formation from the presence of a bed boundary. Thus the setting of parameters to detect skip might also remove valid measurements across thin beds. An alternative approach therefore is to use a clustering algorithm which assumes that valid curves, although they may be affected by small random or systematic errors, generally tend to cluster together and be continuous over depth; and that in general the number of valid curves is greater than the number of invalid curves (or more specifically that valid curves form larger clusters than invalid ones).

Figure 4 provides a high level overview of the clustering algorithm.

1. Transit times from the DFAD algorithm (10) are estimated in the manner described above. Slownesses are computed from these transit times for both receiver and transmitter modes for both transmitters. This pre-processing (12) may also incorporate some data filtering such as removal of suspect measurements as indicated by DFAD status information. 2. A number of depth matched slowness channels over a predetermined depth interval are input (14) to a selection algorithm (16) that classifies them as either valid or invalid. These channels may correspond to either (or both) receiver mode or transmitter mode processing of measured transit times from any transmitter in the tool and from receiver pairs of different spacings.

3. The valid curves are examined on a level by level basis to reject any remaining outliers (identified as being too many standard deviations from the median value) (18).

4. The remaining valid curves are matched to a list of preferences (20) to compute a final result (22). If several matches are found for the list an average will be computed.

5. If no match is found in the first list of preferences (20#1), an attempt can be made to find a match in a second (20#2), third, and so on. If all lists are exhausted without finding a match, an "absent value" may be output. In this preference selection, the first list may correspond to a collection of slownesses needed to make a BHC measurement, the second to a DDBHC, a later one to a single receiver uncompensated measurement, and so on. In this way the best estimate can be computed if possible. If not, the next most desired alternative is selected.

6. Although the clustering algorithm described above is presented in terms of computing a single output (typically a 6" receiver spacing slowness), the possibility exists to provide multiple outputs (e.g., a 6" spacing slowness and a 2ft spacing slowness). The cluster analysis would verify that both outputs were of acceptable quality and consistent.

7. A post-processor may also be added to generate quality control indicators such as number of inputs rejected, which preference generated the output, or a signal to noise estimate for the selected curves based on DFAD data.

Figure 5 shows the operation of a data processing scheme incorporating methods according to the invention in more detail.

The top row (I) corresponds to the pre-processing stage in which slowness are computed from input transit times. Since the clustering algorithm may be demanding on processing power, and since it may not always be required if input data quality is good, provision may be made for an alternative, light processing, evaluating the result, and only continuing with the full algorithm when it is necessary. The Hodges-Lehman averaging approach described above is applicable here. The centre row (II) corresponds to processing of curves over a finite depth interval. The purpose is to identify curves that are relatively similar over such an interval. Curves that skip frequently between good and bad values, and curves that may be stable, but consistently different from the others will be eliminated.

The bottom row (III) shows the steps involved in level-by-level processing. Curves subject to a large number of detection problems will have been rejected in the previous step. However, curves with infrequent skips will remain. These residual skips are first removed, then the output value is computed using the remaining slownesses selected according to one or more lists of preferences. A quality indicator may also be generated.

(I) Pre-processing

The DFAD (Digital First Arrival Detection) algorithm outputs an estimated arrival time (100), a status word (102), the amplitude of the signal peak at the point of detection and a measurement of the noise in the signal before the detection window, as described above. A transit time pre-processing is performed (104) using the DFAD status information to suppress computation of slownesses from transit times that are not reliable. The DFAD algorithm can output a "transit time repeated" indicator where it was unable to determine a transit time and so has output the same value as previously output. Since such an output brings no new information, it is removed from the processing.

Slownesses are computed from the transit times for pairs of receivers (106) taking as a further input the tool geometry for the particular pair in question (108).

Extreme values of slowness are established (1 10) Smin (e.g. 40μs/ft) and Smax (e.g. 200μs/ff) and used to eliminate slownesses falling outside known physical limits of the system (1 12). What remains is a set of raw slownesses (1 14). These can optionally be applied to a fast processing scheme ( 1 16) such as the Hodges-Lehamn averaging approach described above. If the results of such fast processing are considered to be acceptable (e.g. slowness is substantially continuous over the interval in question and shows known formation features), the slowness can be output (1 18). If the result of the fast processing is considered unsatisfactory, or if fast processing is not performed, the data is applied to depth interval curve processing (II).

(II) Depth Interval Processing The raw slownesses from the pre-processing stage are presented as slowness curves over the depth interval of interest. Filtering and resampling (120) of slowness curves is the first step in the process of identifying curves that correlate over a finite depth interval. When slowness with different depth resolutions (i.e., with different receiver spacings) are to be compared, it is necessary to filter them to achieve a common resolution and a comparable response in thin beds. The filtering reduces all of the data to relate to a common measurement resolution, i.e. a notional common receiver spacing.

Once the data has been filtered, high resolution sampling is no longer necessary. Therefore the data may be decimated in order to reduce the number of computations required for processing. A sample interval of half the filter length should be adequate for this purpose

Slowness curve separation is then computed from the filtered and resampled curves (122). The average slowness offset between each filtered curve and all of the others is computed. The average slowness offset or separation between slowness curves /^' and j is defined as if ill Infill ^s,_j = ∑ |ΔTtn, ] — Δ7^*[ ι, j ] / ∑ 1 AT[n,i] ≠ absent , AT[n,j] ≠ absent, j > i n=l / n=l

where DT[n,i] denotes the value of slowness curve # at filtered depth index, n, and the filtered curve consists of nfilt samples. (Note that if there are N slownesses being processed, then (N -N)/2 separation values will be computed.)

Groups (124) are formed of slowness curves whose separation is small (below a specified threshold) or whose separation from a common third curve is small: curves, i & j e group n, if S <SepLim or i,k e group n, j,k e group m In certain circumstances we may have a good idea of the expected formation slowness ΔT_guιde (127) and we wish to ensure only groups around this slowness are accepted. An example might be when STC (slowness time coherence processing) has been run to obtain a standard resolution slowness log, but we wish to reprocess arrival time measurements to create a high resolution log. In this case we compute an average separation between the curves in each group and the guide slowness. The separation between the guide slowness, AT _de and group # , containing groupsize[m] member curves, is given by

S„[m] = 1

Curves in any group, m, where S_gψ[m] > ΔTSepLim (125) are removed from the list of valid slowness curves (126).

Finally groups of a size which is small relative to the size of the largest group found are likely to result from correlations of errors. Thus a test is made on relative group size. The curves in any group, m, where groupsize_va <GrpSizeLim-m-\(groupsize) (125) are removed from the list of valid slowness curves (126).

Ideally, when all or most of the slownesses correlate, only one group will be found. If no groups are found, no curves correlate well enough to be considered reliable, so no output can be computed. If a large number of small groups are found, then further processing may also be difficult since it is difficult to decide which may be good and which bad. In this case also, no output may be computed.

However data may be partitioned into two, three of four groups for valid physical reasons. If borehole effects (mud path length differences) are present then receiver mode measurements from a transmitter above the receivers and transmitter mode slownesses from a transmitter below the receivers will be biased in one direction, while receiver mode measurements from a transmitter below the receivers and transmitter mode slownesses from a transmitter above the receivers will be biased in the opposite direction. If formation alteration has occurred then measurements with short transmitter to receiver spacing will tend to be slower than measurements with long transmitter to receiver spacing.

Each of the slowness curves input to the analysis may be assigned a near/far status, NF, and a BHC status, BHC. A slowness measurement, /, is assigned a near/far status of NF[ ] = 1 if the distance of the transmitter to the mid-point of the receivers is less than or equal to a given threshold, NearFarSpcLim, else it is zero. When borehole effects are present the "polarity"' of the effect is opposite for transmitters above and below the receivers. A slowness measurement, , is assigned a BHC status of BHC[i] = 1 if it is a receiver mode measurement and the transmitter is above the receivers, or if it is a transmitter mode measurement and the transmitter is below the receivers, else it is zero.

For a group of curves the near/far status of the group, grpNF, is simply the sum of the near/far status of the member curves of the group divided by the number of members. ngroιφ[i] grpNF[i] = ∑ NF[n]/ngroup[i] where ngroup[i] is the size of group, /^'.

Similarly, grpBHC[i}

When two groups are present and they are not unreasonably different in slowness, a check is made to see if they correspond to a partitioning of data due to alteration or borehole effect. If the presence of two groups can be explained by formation alteration, then one group will have a near/far status close to 1 and the other close to zero. The expression

- grpNF[2]\ > GrpNFLim

should test true even when GrpNFLim ~ 1. Similarly the presence of borehole effects is indicated when \grpBHC[l] - grpBHC[2]\ > GrpBHCLim is true. If three or four groups are found it is necessary to test whether both alteration and borehole effects are acting simultaneously. Group types are defined as shown in Figure 6. If both group near/far status and BHC status are close to 0, then the group is of type 1 ,..., if both are almost 1 then the group is of type 4. Groups which cannot be assigned to types 1 to 4 are by default classed as type zero. If the presence of multiple groups is to be explained as combined alteration and borehole effects, then each group should have a different type, and no group should be of type 0. In addition the separation between all groups should not be unreasonably large. Figure 7 shows a flow diagram of a method applied to two to four groups to provide a single group of curves for further processing.

If it can be shown that multiple groups may be the result of alteration or borehole effects, then it may be concluded that they all contain valid data. So they may be combined and treated as a single group in further, level-by-level processing.

(III) Level-by-Level Processing

The goal of the formation of groups described above is to identify slownesses that correlate on average well with each other. These slownesses have the greatest probability of being correct on average. However it is probable that individual measurements may still be affected by false detection errors: either cycle skips or noise problems. It is therefore necessary to make a pass through these data, on a level by level basis to identify such problems and remove the affected data points from the slowness curves that have been retained by the preceding steps. The method selected needs to be flexible enough to deal correctly with thin beds where curves of differing resolution will read differently, and with alteration and borehole effects where slownesses may differ between curves. The method suggested here is to use the standard deviation of the measurements about the median as a test for outliers (128).

The standard deviation about the median slowness, ΔT_medmι is defined as

Slowness values that fail the test

\AT[n] - AT_medmn I < SdAcceptLim ^■ SD_AT are rejected ( 130).

When a slowness is rejected, the standard deviation of the remaining values is recomputed, and the remaining values are compared to this new standard deviation. In this way, large cycle skips will not prevent the rejection of smaller detection errors. When the majority of slowness curves cluster about some value, the standard deviation about the median will be small and outliers will tend to lie several standard deviations from the median. When there is considerable variation among valid curves (e.g., in thin beds) the standard deviation will increase to encompass the natural spread of the data.

The selection of the test parameter, SdAcceptLim, which determines how many standard deviations a point may depart from the median and still be considered valid, needs to be chosen with care. There is a trade off between rejecting good data when there is a spread for physical reasons, and failing to reject detection problems when the error is small. A value of 2 1/2 to 3 standard deviations is probably a good starting point (132).

At the end of processing a list of slownesses (134) which are considered valid at each depth level of the processing interval has been created ( 136). One or more outputs (1 18) may now be computed (142)from the members of this list. Fig. 8 shows an example to illustrate how this may be done. For the sake of simplicity only two preferences are presented. Preference #1

(200) consists of four pairs of entries of the form "TlRlR2r T2Rl R2r", "TlR2R3r T2R2R3r".

The code, "TlRlR2r", is interpreted as the slowness computed for a firing of transmitter #1 using transit times of receivers #1 and #2 in receiver mode. If all slownesses specified on a line of a preference list are available in the list of valid slownesses at a given depth, then an output

(205) is produced as the linear combination of the corresponding slownesses, using the coefficients from the corresponding "preference coefficient list" (203). If the example shown here is applied to the tool of Figure 1 , each pair on a line of preference 1 corresponds to a slowness from the upper transmitter and one from the lower transmitter on the same pair of receivers. The corresponding coefficients are both 0.5, so the result is to compute the average of the two slownesses: a BHC slowness value. If any entry on a line cannot be matched with a slowness in the valid group, then no output value is computed for that line. If output values are computed for more than one line in a preference list, then these are averaged to provide the final output (206).

If no output is produced for any line in a preference list, then the next list in the series is processed. For the example of Fig. 8, the second preference list (202) contains entries of the form "TlRlR2r TlRlR2x". As explained above the first entry denotes the slowness computed for a firing of transmitter #1 using transit times of receivers #1 and #2 in receiver mode. The second denotes the transmitter mode slowness from the same transmitter and receiver pair. Combining these with the coefficients of 0.5 from the associated preference coefficient list (204) would result in a DDBHC (depth derived borehole compensated) slowness value.

This example shows how it is possible to output a BHC measurement if all required data is available. If, for example, transmitter #2 happened to fail then the algorithm would automatically switch to the next best alternative, such as a DDBHC measurement based only on data from transmitter #1. Further lists can be added to cater for additional fall back solutions for other problems. (Other coefficient values could also be used to combine slowness values in ways to create outputs other than slowness. For example estimates of borehole or alteration effects may be derived by the appropriate choice of coefficient values.)

The following output curves (148) may be generated (144) to provide some idea of the quality of the measurements input to the slowness processing algorithm: - fraction of total number of input slownesses retained as valid,

- identity of preference list actually used to produce the output slowness, (i.e., whether first preference, second, or so on...),

- fraction of number of curves in preference list used actually contributing to output slowness,

- if DFAD noise and peak amplitude measurements are available (146), signal to noise ratio, based on transit time measurements actually contributing to the output slowness. Code to perform the processing steps described above can be readily implemented in MATLAB.

INDUSTRIAL APPLICABILITY The present invention finds application in the field of acoustic logging tools which can be used to evaluate the formations surrounding boreholes such as are drilled for the extraction of hydrocarbons or geothermal energy.

Claims

1 A method of processing data obtained from an interval of a borehole using an array tool including multiple receivers to make a series of measurements in the interval, the method comprising:

(i) determining from the data, a parameter relating to the formation around the borehole for each pair of receivers in the array for each measurement in the series; (ii) determining comparisons of pairs of parameters relating to measurements at the same location in the interval; and (iii) processing statistically the comparisons to determine an acceptable value for the parameter at each location in the interval.

A method as claimed in claim 1 , further comprising removing any determined parameters which fall outside predetermined limits based on known physical properties of the borehole prior to determining comparisons of pairs of parameters relating to measurements at the same location in the interval.

A method as claimed in claim 2, wherein the data is sonic logging data.

A method as claimed in claim 3, wherein the parameter is related to the sonic slowness of the formation.

A method as claimed in claim 3 or 4, wherein the tool comprises two transmitters and an array of receivers.

A method as claimed in claim 5, wherein the parameter comprises the difference in signal transit times measured on any given pair of receivers for a signal originating with one or other transmitter (receiver mode). A method as claimed in claim 5 wherein the parameter comprises the difference in signal transit times measured on any given pair of receivers for signals originating with each transmitter when at a given location in the interval (transmitter mode).

A method as claimed in any of claims 3 to 7, comprising the steps of:

(a) computing formation slowness curves for each pair of receivers over the interval;

(b) determining upper and lower thresholds and eliminating any values of slowness lying outside these thresholds;

(c) determining the separation of each curve from the other curves for the interval; (d) forming groups of curves based on the determined separations;

(e) selecting one or more groups with the highest number of members;

(f) processing the data in the or each group statistically to remove outliers;

(g) determining the formation slowness from the processed group.

A method as claimed in any of claims 3 to 7, comprising the steps of:

(a) determining transit time differences for each pair of receivers in the array;

(b) determining upper and lower thresholds and eliminating any transit time differences lying outside these thresholds; and

(c) determining the Hodges-Lehman average of the remaining transit times to produce an slowness curve for the interval.

A method as claimed in any of claims 3 to 7, comprising the steps of:

(c) determining the median of the remaining transit times to produce a slowness curve for the interval.

A method as claimed in claim 9 or 10, further comprising the steps of: (i) determining values of a parameter related to formation slowness for each pair of receivers separated by a predetermined distance in both receiver mode and transmitter mode;

(ii) eliminating values lying outside the thresholds; (iii) determining an offset in the variation of parameter value with transmitter to receiver spacing;

(iv) applying the offset to the parameter values and removing any values falling outside the thresholds;

(v) determining the Hodges-Lehman average of the remaining values by computing the average values of all combinations of pairs of receivers and determining the median value of the result.

A method as claimed in claim 8, 9, 10 or 1 1 , further comprising the step of outputting a quality indicator at each measurement location in the interval.

A method as claimed in any of claims 3 to 12, wherein the output data is selected to provide a borehole compensated result.